ClawMorph

Turn one agent into many roles — with preview, snapshot, and rollback.

Transform any OpenClaw agent into a role-specific workspace in one command.

ClawMorph lets you turn the same agent into a researcher, designer, lawyer, product manager, or founder without hand-editing prompt files.

Unlike simple prompt switchers, ClawMorph works at the workspace layer:

• preview changes before writing anything
• apply a role pack safely
• snapshot the previous state automatically
• roll back cleanly when you want to undo

That makes it useful for real agent iteration, not just prompt swapping.

Why people star it

• Role switching without manual file surgery
• Safer than ad-hoc prompt edits because every apply is previewable and reversible
• Works on real OpenClaw workspaces instead of isolated prompt snippets
• Easy to demo: list, preview, apply, rollback

Quick demo

npm install
npm run build
npm run list
npm run preview -- --path ./test-fixtures/demo-agent --role researcher
npm run demo:apply
npm run demo

npm run demo:apply uses a temporary copy of the demo fixture so you can test apply + rollback safely without mutating the checked-in files.

Why it is different

| Capability | Prompt template | ClawMorph | |---|---:|---:| | Switch role instructions | ✅ | ✅ | | Preview changes before write | ❌ | ✅ | | Update workspace files | ❌ | ✅ | | Snapshot before mutation | ❌ | ✅ | | Roll back to previous state | ❌ | ✅ |

Before vs after

Before

A generic OpenClaw agent workspace might only contain a light identity file and a few loose notes.

After

ClawMorph can apply a role pack that:

• updates IDENTITY.md
• creates or updates SOUL.md
• creates or updates TOOLS.md
• appends role-specific entries to MEMORY.md
• stores a snapshot in .clawmorph/snapshots/...

Built-in role packs

• researcher — evidence gathering, source comparison, synthesis
• designer — UX reasoning, flows, visual direction
• lawyer — careful wording, constraints, risk analysis
• product-manager — scope, priorities, outcomes, trade-offs
• founder — strategy, momentum, business judgment

Commands

Create a new agent workspace

npm run new -- my-agent --role researcher
npm run new -- founder-bot --role founder --root /tmp/clawmorph-agents

This creates a new OpenClaw-compatible workspace and immediately applies the selected role pack.

List available role packs

npm run list
npm run snapshots -- --path /tmp/leo-demo

Preview changes for a role pack

npm run preview -- --path ./test-fixtures/demo-agent --role researcher
npm run preview -- --path ./test-fixtures/demo-agent --role researcher --json

Or resolve a named agent conservatively:

npm run preview -- --agent leo --role researcher

Apply + rollback safely on a temporary copy

npm run demo:apply

This runs apply and rollback against a temporary copy of test-fixtures/demo-agent.

Apply a role pack directly to a workspace copy

npm run apply -- --path /tmp/leo-demo --role researcher

Roll back the latest apply

npm run rollback -- --path /tmp/leo-demo
npm run rollback -- --path /tmp/leo-demo --snapshot <snapshot-id>
npm run demo

How it works

ClawMorph follows a simple flow:

1. Resolve target workspace
   • explicit --path
   • or a best-effort --agent lookup
2. Load a role pack from role-packs/*.yaml
3. Scan the workspace for key OpenClaw files
4. Preview planned changes
5. Create a snapshot before writing files
6. Apply file updates
7. Rollback if needed

Current MVP focuses on these files:

• IDENTITY.md
• SOUL.md
• TOOLS.md
• MEMORY.md

ClawMorph intentionally does not mutate USER.md in this version.

Automation-friendly output

For integration with scripts or future UI layers, core commands support JSON output:

npm run list -- --json
npm run preview -- --path ./test-fixtures/demo-agent --role researcher --json
npm run apply -- --path /tmp/leo-demo --role researcher --json
npm run rollback -- --path /tmp/leo-demo --json
npm run snapshots -- --path /tmp/leo-demo --json

Safe real-agent demo

If you want to test against a real OpenClaw agent without mutating the original workspace, copy that agent directory first and run ClawMorph against the copy.

Example:

cp -R ~/.openclaw/workspace/agents/leo /tmp/leo-demo
npm run preview -- --path /tmp/leo-demo --role researcher
npm run apply -- --path /tmp/leo-demo --role researcher
npm run rollback -- --path /tmp/leo-demo

This keeps your real agent untouched while still demonstrating the workflow.

Cross-platform note

ClawMorph's core CLI is Node.js-based and is intended to work across macOS, Ubuntu/Linux, and Windows.

• Core commands (list, preview, apply, rollback) are implemented in TypeScript/Node and are platform-friendly.
• Demo entrypoints now use Node-based scripts instead of Unix-only shell helpers.
• For the smoothest first run on Windows, prefer the explicit --path mode before testing --agent.

Current scope

Included in this MVP

• role-pack listing
• workspace scanning
• preview planning
• apply with snapshots
• rollback of latest snapshot

Not included yet

• web UI
• plugin packaging
• OpenClaw skill wrapper
• role marketplace / registry
• team features
• cloud sync

Roadmap

Next likely improvements

• better README visuals / GIF demo
• improved real-agent path resolution
• cleaner generated sections and formatting
• import/export role packs
• publishable npm package workflow
• optional OpenClaw skill wrapper around the CLI

Development

npm install
npm run build
npm run list
npm run preview -- --path ./test-fixtures/demo-agent --role researcher

License

MIT